Pattern inspection method and its apparatus

This application claims priority to Japanese Patent Application No. 2003-065886 filed on Mar. 12, 2003.

BACKGROUND OF THE INVENTION

The present invention relates to an inspection method and apparatus for comparing an image of an object, which is obtained by using light, laser beams, or the like, and a reference image to detect a fine pattern defect, a foreign body, or the like from a difference between the images. In particular, the present invention relates to a pattern inspection apparatus that is designed preferably for performing visual inspection of a semiconductor wafer, a TFT, a photomask, and the like, and a method therefor.

As an example of a conventional technique for comparing an inspection object image and a reference image to detect a defect, reference is made to a method described in JP-A-05-264467. This method involves sequentially sensing images of an inspection object specimen using a line sensor, in which repetitive patterns are arranged regularly, comparing the sensed images with images delayed by a time for establishing a repetitive pattern pitch, and detecting a non-coincident part of the images as a pattern defect.

Such a conventional inspection method will be described in conjunction with the visual inspection of a semiconductor wafer, as an example. In a semiconductor wafer which serves as an object of inspection, as shown in <figref idref="DRAWINGS">FIG. 6</figref>, a large number of chips having an identical pattern are arranged regularly. As shown in <figref idref="DRAWINGS">FIG. 7</figref>, each chip can be roughly divided into a memory mat section <b>71</b> and a peripheral circuit section <b>72</b>. The memory mat section <b>71</b> consists of a set of small repetitive patterns (cells), and the peripheral circuit section <b>72</b> basically consists of a set of random patterns. In general, in the memory mat section <b>71</b>, the pattern density is high, and an image thereof obtained by a bright-field illumination optical system tends to be dark. On the other hand, in the peripheral circuit section <b>72</b>, the pattern density is low, and an image thereof tends to be bright.

In the conventional visual inspection, images in the same positions of two chips adjacent to each other, for example, an area <b>61</b>, an area <b>62</b>, and the like, as seen in <figref idref="DRAWINGS">FIG. 6</figref>, are compared, and a difference between the images is detected as a defect. In this case, since there may be a vibration of the stage which holds the object, inclination of the object, or the like, the positions of the two images do not always coincide with each other. Thus, the amount of positional, deviation of the image sensed by the sensor and the image delayed by the repetitive pattern pitch is determined, the two images are aligned on the basis of the amount of positional deviation, and then a difference between the images is calculated. When the difference is larger than a specified threshold value, it is judged that there is a defect in the pattern; and, when the difference is smaller than the threshold value, it is judged that there is no defect in the pattern.

In the alignment of two images in the comparative inspection, it is a general practice to set the edge parts in the images as one piece of information for calculation of the amount of positional deviation and to calculate the amount of positional deviation such that deviation of the edge parts between the images is minimized. Actually, a method has been proposed using normalized cross correlation, as well as a method using a sum of residuals, and the like. However, in any of such methods, since the amount of calculation is enormous, in order to realize speedup of the inspection, various measures have been required, such as changing the positional deviation calculation section to hardware or increasing the number of arithmetic operation circuits, and change of the image processing algorithm, such as simplification of the calculation of the amount of positional deviation.

In addition, in a semiconductor wafer which serves as the object of inspection, a slight difference in thickness occurs in a pattern due to planarization by CMP or the like, and so there is a difference in the brightness locally in images among chips. For example, reference numeral <b>41</b> in <figref idref="DRAWINGS">FIG. 4A</figref> denotes an example of an inspection object image, and reference numeral <b>42</b> in <figref idref="DRAWINGS">FIG. 4B</figref> denotes an example of a reference image. As indicated by the pattern <b>4</b><i>a </i>in <figref idref="DRAWINGS">FIG. 4A</figref> and the pattern <b>4</b><i>b </i>in <figref idref="DRAWINGS">FIG. 4B</figref>, a difference in the brightness occurs in an identical pattern of the inspection object image and the reference image.

In addition, there is a defect <b>4</b><i>d </i>in the inspection object image <b>41</b> of <figref idref="DRAWINGS">FIG. 4A</figref>. A difference image in this case is as shown in <figref idref="DRAWINGS">FIG. 4C</figref>. The difference image is an image represented by a density difference according to a differential value in corresponding positions of an inspection object image and a reference image. A waveform of a differential value in position <b>1</b>D-<b>1</b>D′ is as shown in <figref idref="DRAWINGS">FIG. 4D</figref>. With respect to such an image, if a part where the differential value is equal to or more than a specific threshold value TN is regarded as a defect, as in the conventional system, a differential value <b>4</b><i>c </i>of the patterns <b>4</b><i>a </i>and <b>4</b><i>b</i>, which are different in brightness, is detected as a defect. However, this condition should not originally be detected as a defect. In other words, this is a false defect or a nuisance defect (hereinafter referred to as a false defect). Conventionally, as one method of avoiding the occurrence of a false defect, such as indicated by the differential value <b>4</b><i>c </i>of <figref idref="DRAWINGS">FIG. 4C</figref>, the threshold value TN is increased (from TH to TH<b>2</b> in <figref idref="DRAWINGS">FIG. 4D</figref>). However, this leads to a decrease in the sensitivity, and the defect <b>4</b><i>d </i>with a differential value of the same or lower level cannot be detected.

In addition, a difference in the brightness due to a difference in the thickness may occur only among specific chips within a wafer such as shown in <figref idref="DRAWINGS">FIG. 6</figref>, or it may occur only in a specific pattern within a chip. However, if the threshold value TH is adjusted to these local areas, the overall inspection sensitivity will be extremely decreased.

SUMMARY OF THE INVENTION

The present invention is directed to a pattern inspection apparatus for comparing images of corresponding areas of two patterns, which are formed so as to be identical, to judge that a non-coincident part of the images is a defect, and serves to reduce the influence of unevenness in brightness of patterns caused by a difference in the thickness or the like, so as to realize a highly sensitive pattern inspection. In addition, the present invention realizes high-speed pattern inspection without the need for changing the image comparison algorithm. Consequently, the problems of the conventional inspection technique can be solved.

In other words, in accordance with the present invention, a pattern inspection apparatus for comparing images of corresponding areas of two patterns, which are formed so as to be identical, so as to judge that a non-coincident part of the images is a defect, includes means for performing processing for detecting an image with an image sensor, processing for comparing images, and processing for judging a defect at different times. As a specific example of such means, the pattern inspection apparatus includes means for performing processing for comparing images in parallel in plural areas. Therefore, the pattern inspection apparatus includes plural units for performing processing for comparing images and a number of memories equal to or less than the number of units. In addition, the pattern inspection apparatus includes means for writing data in a memory of the image sensor and reading out data from the memory at different times.

Further, the pattern inspection apparatus includes means for converting gradation of an image signal among compared images by different plural processing units. Consequently, even in the case in which an object of inspection is a semiconductor wafer and a difference in brightness occurs in an identical pattern among images due to a difference in thickness in the wafer, or where there is a fluctuation in the quantity of illumination light and a fluctuation in the sensitivity for each pixel of the image sensor, or unevenness in the quantity of light accumulation time, a defect can be detected correctly.

These and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

In the accompanying drawings:

<figref idref="DRAWINGS">FIG. 1</figref> is a block diagram showing an example of the structure of an inspection apparatus according to the present invention;

<figref idref="DRAWINGS">FIG. 2</figref> is an image diagram showing an example of the flow of image data in the course of parallel processing by two image comparison processing units;

<figref idref="DRAWINGS">FIG. 3</figref> is an image diagram showing another example of the flow of image data in the course of parallel processing by two image comparison processing units;

<figref idref="DRAWINGS">FIG. 4A</figref> is a diagram showing an inspection object image at the time when there is unevenness in brightness among compared chips in an example of a conventional threshold value setting method;

<figref idref="DRAWINGS">FIG. 4B</figref> is a diagram showing a reference image in the conventional threshold value setting method;

<figref idref="DRAWINGS">FIG. 4C</figref> is a difference image in the conventional the threshold value setting method;

<figref idref="DRAWINGS">FIG. 4D</figref> is a waveform diagram of a differential value in position <b>1</b>D-<b>1</b>D′;

<figref idref="DRAWINGS">FIG. 5</figref> is a plan view of a semiconductor wafer and shows an enlarged view of an example of an image comparison processing unit;

<figref idref="DRAWINGS">FIG. 6</figref> is a plan view of a semiconductor wafer and shows an enlarged view of a chip;

<figref idref="DRAWINGS">FIG. 7</figref> is a plan view showing an example of the structure of a chip;

<figref idref="DRAWINGS">FIG. 8</figref> is a flow diagram showing a flow of processing for adjusting a difference in brightness which occurs in a belt shape;

<figref idref="DRAWINGS">FIG. 9</figref> is a flow diagram showing a flow of processing in the image comparison processing unit;

<figref idref="DRAWINGS">FIG. 10A</figref> is a diagram illustrating processing for adjusting a difference in brightness which occurs in a belt shape and shows an example of adjusting the brightness in a row direction;

<figref idref="DRAWINGS">FIG. 10B</figref> is a diagram illustrating processing for adjusting a difference in brightness which occurs in a belt shape and shows an example of adjusting the brightness in a column direction;

<figref idref="DRAWINGS">FIG. 10C</figref> is a diagram illustrating processing for adjusting a difference in brightness which occurs in a belt shape and shows an example of adjusting the brightness in a row direction and a column direction;

<figref idref="DRAWINGS">FIG. 11A</figref> is a diagram showing an example of a result of processing by a brightness correction unit and shows a difference image after alignment;

<figref idref="DRAWINGS">FIG. 11B</figref> a diagram showing an example of a result of processing by the brightness correction unit and shows an image in which a difference in brightness is corrected at a high frequency (by a unit of linear or belt-like area);

<figref idref="DRAWINGS">FIG. 11C</figref> a diagram showing an example of a result of processing by the brightness correction unit and shows an image in which a difference in brightness is corrected at a low frequency (by a unit of a wide area);

<figref idref="DRAWINGS">FIG. 11D</figref> is a diagram showing an example of a result of processing of the brightness correction unit and shows a luminance waveform diagram along line <b>1</b>D-<b>1</b>D′ in <figref idref="DRAWINGS">FIG. 11A</figref>;

<figref idref="DRAWINGS">FIG. 11E</figref> is a diagram showing an example of a result of processing by the brightness correction unit and shows a luminance waveform diagram of <figref idref="DRAWINGS">FIG. 11C</figref> corresponding to line <b>1</b>D-<b>1</b>D′ of <figref idref="DRAWINGS">FIG. 11A</figref>;

<figref idref="DRAWINGS">FIG. 12</figref> shows an example of a proper threshold value according to coordinates in a wafer and coordinates in a chip in a plan view of a semiconductor wafer;

<figref idref="DRAWINGS">FIG. 13</figref> is a flow diagram showing a flow of processing for setting a threshold value for each of the coordinates in a wafer and the coordinates in a chip;

<figref idref="DRAWINGS">FIG. 14</figref> is a plan view of a semiconductor wafer showing an example of inspection chip setting at the time of trial inspection;

<figref idref="DRAWINGS">FIG. 15</figref> is a front view of a displayed screen showing an example of superimposed indication of design data of a chip and processing results;

<figref idref="DRAWINGS">FIG. 16A</figref> is a graph showing an example of setting a threshold value from characteristic amounts of a defect and a false defect and shows a relation between a density difference value and the frequency of the defect and the false defect;

<figref idref="DRAWINGS">FIG. 16B</figref> is a graph showing an example of setting a threshold value from characteristic amounts of a defect and a false defect and shows a relation between an area and a density difference value of the defect and the false defect;

<figref idref="DRAWINGS">FIG. 17</figref> is a flow diagram showing a flow of processing for adjusting a difference of brightness which occurs at random;

<figref idref="DRAWINGS">FIG. 18A</figref> is a diagram showing a reference image, a detected image, and a difference image for which the brightness by a unit of belt-like area is adjusted;

<figref idref="DRAWINGS">FIG. 18B</figref> is a graph showing a characteristic amount space which is formed by the reference image and the detected image;

<figref idref="DRAWINGS">FIG. 18C</figref> is a scatter diagram in a characteristic amount space with a brightness of the detected image plotted on the X axis and a brightness of the reference image plotted on the Y axis;

<figref idref="DRAWINGS">FIG. 18D</figref> shows scatter diagrams in which the scatter diagram of <figref idref="DRAWINGS">FIG. 18C</figref> is divided for each characteristic amount;

<figref idref="DRAWINGS">FIGS. 19A and 19B</figref> are diagrams showing an example of a segment dividing method for a characteristic space; and

<figref idref="DRAWINGS">FIG. 20</figref> is a flow diagram showing a processing procedure for calculating a divided threshold value for the segment division.

DESCRIPTION OF THE PREFERRED EMBODIMENT

An embodiment of the present invention will be described in detail with reference to <figref idref="DRAWINGS">FIGS. 1 to 20</figref>.

As an example of the present invention, a defect inspection method employed in an optical visual inspection apparatus targeting a semiconductor wafer will be considered. <figref idref="DRAWINGS">FIG. 1</figref> shows an example of the structure of the apparatus. Reference numeral <b>11</b> denotes a specimen (an object to be inspected, such as a semiconductor wafer). The apparatus includes a stage <b>12</b> on which the specimen <b>11</b> is placed and moved; and a detection unit <b>13</b>, including a light source <b>101</b>, which illuminates the specimen <b>11</b>, an illumination optical system <b>102</b>, which condenses light emitted from the light source <b>101</b>, an object lens <b>103</b>, which illuminates the specimen <b>11</b> with the illumination light condensed by the illumination optical system <b>102</b> and focuses an optical image obtained by reflection of the illumination light, an image sensor <b>104</b>, which receives the focused optical image and converts the focused optical image into an image signal corresponding to a value of brightness, and an A/D conversion unit <b>105</b>, which converts an input signal received from the image sensor <b>104</b> into a digital signal.

Here, a lamp is used as the light source <b>101</b> in the example shown in <figref idref="DRAWINGS">FIG. 1</figref>. However, a laser may be used. In addition, light emitted from the light source <b>101</b> may be light of a short wavelength, or it may be light of a wavelength in a wide band (white light). In the case in which light having a short wavelength is used, in order to increase the resolution of an image to be detected (to detect a fine defect), light having a wavelength in the ultraviolet region (ultraviolet light: UV light) may be used.

In addition, it is possible to detect a defect at a relatively high speed and with a high sensitivity by adopting a time delay integration image sensor (TDI image sensor), which is constituted by arranging plural one-dimensional image sensors two-dimensionally, as the image sensor <b>104</b>, to transfer a signal detected by each one-dimensional image sensor to the one-dimensional image sensor in the next stage in synchronization with movement of the stage <b>13</b> and to add up the signals.

The apparatus also includes an image editing unit <b>14</b>, which includes a pre-processing section <b>106</b>, which applies image correction, such as shading correction and dark level correction, to a digital signal of an image detected in the detection unit <b>13</b>, and image memories <b>107</b>-<b>1</b> and <b>107</b>-<b>2</b> in which digital signals of a detected image to be an object of comparison and a reference image are stored.

The apparatus further includes an image comparison processing unit which calculates defect candidates in a wafer used as a specimen. The apparatus uses plural image comparison processing units (<b>15</b>-<b>1</b>, <b>15</b>-<b>2</b>). The image comparison processing unit <b>15</b>-<b>1</b> compares two images (a detected image and a reference image) stored in an image memory <b>107</b>-<b>1</b> of the image editing unit <b>14</b> and judges that a part where a differential value is larger than a threshold value is a defect. First, the image comparison processing unit <b>15</b>-<b>1</b> reads out digital signals of the detected image and the reference image stored in the image memory <b>107</b>-<b>1</b>, calculates an amount of positional deviation for adjusting the deviation of positions in a positional deviation detection section <b>108</b>-<b>1</b>, and calculates an amount of signal correction for adjusting the deviation of the brightness between the detected image and the reference image in a brightness correction section <b>109</b>-<b>1</b>. Then, the image comparison processing unit <b>15</b>-<b>1</b> compares corresponding positions of the detected image and the reference image in brightness in an image comparison section <b>110</b>-<b>1</b> using the calculated amount of positional deviation and amount of signal correction, and identifies a part where a differential value is larger than a specific threshold value as a defect candidate. A threshold value setting section <b>111</b>-<b>1</b> sets a threshold value, which is used in extracting a defect candidate from a differential value for each area and gives the threshold value to the image comparison section <b>110</b>-<b>1</b>.

In the image comparison processing unit <b>15</b>-<b>2</b>, processing is performed according to the same procedure as the image comparison processing unit <b>15</b>-<b>1</b> using two images stored in an image memory <b>107</b>-<b>2</b>.

The apparatus has an overall control unit <b>16</b>, including a user interface section <b>112</b> which provides display means and input means for receiving a change of inspection parameters (a threshold value and the like used in image comparison) from a user and for displaying information on a detected defect, a storage <b>113</b> which stores a characteristic amount of a detected defect candidate, an image, and the like, and a CPU (incorporated in the overall control unit <b>16</b>) which performs various controls. This is a mechanical controller <b>114</b> which drives the stage <b>12</b> on the basis of a control command received from the overall control unit <b>16</b>. Note that the image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b>, the detection unit <b>13</b>, and the like are also driven according to a command from the overall control unit <b>16</b>.

In the semiconductor wafer <b>11</b> which serves as an object of inspection, a large number of chips of an identical pattern are arranged regularly as shown in <figref idref="DRAWINGS">FIG. 6</figref>. In the inspection apparatus of <figref idref="DRAWINGS">FIG. 1</figref>, the semiconductor wafer <b>11</b> is continuously moved by the stage <b>12</b> in response to the overall control unit <b>16</b>, and in synchronization with this, images of the chips are sequentially captured from the detection unit <b>13</b>, digital image signals of the same positions of two adjacent chips, for example, the area <b>61</b> and the area <b>62</b> in <figref idref="DRAWINGS">FIG. 6</figref>, are compared as a detected image and a reference image, respectively, according to the above-mentioned procedure, and a difference between the images is detected as a defect.

Here, the inspection apparatus of this example has the two image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> and performs processing in parallel. In addition, the image memories <b>107</b>-<b>1</b> and <b>107</b>-<b>2</b> also correspond to the respective image comparison processing units. <figref idref="DRAWINGS">FIG. 2</figref> is a diagram showing the flow of processing. Here, a case in which seven chips, namely, chips <b>1</b> to <b>7</b>, are inspected will be described.

First, the image sensor <b>104</b> sequentially captures images of the chips. In association with the image capturing speed of the sensor, the image editing unit <b>14</b> performs pre-processing of the chip images in the pre-processing section <b>106</b>, and the chip images are written in a memory <b>107</b>. In this case, the image of the chip I is written in the memory <b>10</b><b>7</b>-<b>1</b> for the image comparison processing unit <b>15</b>-<b>1</b>, the image of the chip <b>7</b> is written in the memory <b>10</b><b>7</b>-<b>2</b> for the image comparison processing unit <b>15</b>-<b>2</b>, and the images of the chips <b>2</b> to <b>6</b> are written in both the memories <b>107</b>-<b>1</b> and <b>107</b>-<b>2</b>. On the other hand, the image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> read out images from the memories <b>107</b>-<b>1</b> and <b>107</b>-<b>2</b>, respectively, at a speed which is half the writing speed, and when two images are readout, image comparison is performed, such as calculation of an amount of positional deviation, calculation of an amount of signal correction, and extraction of defect candidates in parallel with the reading-out of images. The processing speed of the image comparison also is half the image capturing speed of the sensor. The image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> mutually perform inspection of the chips of the other unit in parallel.

<figref idref="DRAWINGS">FIG. 3</figref> is a diagram showing the flow of another kind of processing. Here, in the time for capturing an image of one chip, the two image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> perform image comparison of an area for a half chip, respectively, (e.g., with respect to a length L of the chip, the image comparison processing unit <b>15</b>-<b>1</b> performs image comparison of a former half L/2 and the image comparison processing unit <b>15</b>-<b>2</b> performs image comparison of a latter half L/2). First, the image sensor <b>104</b> sequentially captures the images of the chips. In association with the image capturing speed of the sensor, the image editing unit <b>14</b> performs pre-processing of the chip images, and it writes the image of the former half L/2 of each chip in the memory <b>107</b>-<b>1</b> for the image comparison processing unit <b>15</b>-<b>1</b> and writes the image of the latter half L/2 of each chip in the memory <b>107</b>-<b>2</b> for the image comparison processing unit <b>15</b>-<b>2</b>. On the other hand, the image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> read out images at a speed which is half the writing speed, respectively, and when images for the two chips are read out, image comparison is performed, such as calculation of an amount of positional deviation, calculation of an amount of signal correction, and extraction of a defect candidate in parallel with the reading-out of the images. As shown in <figref idref="DRAWINGS">FIG. 3</figref>, the processing speed for the image comparison is also half the image capturing speed of the sensor. While an image for one chip is captured, the image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> perform inspection for a half length of the chip, respectively, in parallel.

As described above, in a case in which the speed of the detection processing of an image and the speed of the comparison processing are different, in other words, the inspection speed corresponding to the capturing speed of an image sensor can be realized by causing plural image comparison processing units to perform processing in parallel, even if the processing speed of the image comparison processing unit <b>15</b> is half the image capturing speed for an object chip by the image sensor <b>104</b> and the image editing speed. For example, if the image capturing speed of the image sensor is 1.6 Gpps (pps: pixel per second) at the fastest, when it is converted from a relation of an amount of acquired light or the like, even if the processing capability of the image comparison processing unit <b>15</b> is only 0.8 Gpps, which is half the image capturing speed, it becomes possible to realize an inspection processing speed of 1.6 Gpps by adopting this constitution. Naturally, even in the case in which the speed of the image sensor is higher, it is possible to cope with this speed by constituting the image comparison processing unit <b>15</b> described above with three or more image comparison processing units to cause them to process the acquired image signals in parallel. In addition, it is also possible to cope with a case in which an image capturing range of the image sensor is increased.

In the above-mentioned embodiment, a case in which the image sensor <b>104</b> has a single output is described. However, even if the image sensor <b>104</b> is of a type provided with plural output terminals and outputs plural signals in parallel, signals can be processed in the same manner as the above-mentioned embodiment to perform image processing at a higher speed. In this case, plural signal lines are led out from the image sensor <b>104</b>, and the plural signal lines are connected to plural A/D converters <b>105</b> corresponding thereto, respectively, and outputs from the plural A/D converters <b>105</b> are inputted to the image editing unit <b>14</b> and are processed in the procedure as described above.

Next, the processing in the image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> will be described in detail. First, the detected image signal and the reference image signal to be continuously inputted to the memory <b>107</b> are read out in synchronization with the movement of the stage. Since these image signals of two chips are not signals that relate completely to the same position if the stage is vibrated or a wafer set on the stage is inclined, the amount of positional deviation between the two images is calculated in the positional deviation detection section <b>108</b>. The calculation of the amount of positional deviation is sequentially performed with a specific length in the traveling direction of the stage as one processing unit. Reference numeral <b>51</b>, <b>52</b>, . . . in <figref idref="DRAWINGS">FIG. 5</figref> denote respective processing areas in the case in which a length D (pixel) is treated as one processing unit. This unit processing area will be hereinafter described as a unit.

In this way, the amount of positional deviation is sequentially calculated unit by unit with respect to an image to be inputted, for example, the amount of positional deviation is calculated in the unit <b>51</b> and a unit of an adjacent chip corresponding to the unit <b>51</b>, and then the amount of positional deviation is calculated in the unit <b>52</b> and a unit of an adjacent chip corresponding to the unit <b>52</b>. For the calculation of an amount of positional deviation, there are various methods using normalized cross correlation among images, a sum of density differences among images, a sum of squares of density differences among images, and the like. Any of such methods may be used. Further, the alignment of two images is performed unit by unit on the basis of the calculated amount of positional deviation.

Here, in order to perform highly accurate alignment with respect to an amount of positional deviation calculated unit by unit, in accordance with the present invention, the reliability of the amount of positional deviation calculated in each unit is evaluated (e.g., in the case in which an amount of positional deviation is calculated according to normalized cross correlation, a magnitude of a correlation coefficient). In a unit with low reliability (e.g., small correlation coefficient), the amount of positional deviation calculated in the unit is not used, and the amount of positional deviation is calculated from an amount of positional deviation calculated in a unit with higher reliability before or after (in terms of time) the unit by interpolation or extrapolation. Consequently, it becomes possible to perform alignment following distortion of an image due to vibration of the stage even in the case in which there is only a little pattern information for calculation of the amount of positional deviation of the unit.

On the other hand, in the case in which plural output signal lines are connected to plural A/D converters <b>105</b> from the image sensor <b>104</b>, and outputs from the plural A/D converters <b>105</b> are inputted to the image processing unit <b>14</b>, the calculation of an amount of positional deviation unit by unit is also performed in parallel, and the alignment is also performed in parallel. A unit of an area which is divided in a direction perpendicular to a traveling direction of the stage and in which parallel processing is performed is hereinafter described as a channel. Here, in order to perform highly accurate alignment by a unit of a channel as well following distortion, such as vibration of the stage and fluctuation in magnification of an optical system, it is also possible to extract an amount of positional deviation with high reliability (e.g., with a high correlation coefficient) from plural amounts of positional deviation calculated for each channel, subject the extracted plural pieces of positional deviation information to processing, such as compilation and statistical processing, to calculate an amount of positional deviation of a channel with low reliability.

As an example of the calculation, one amount of positional deviation of a channel with highest reliability among all channels is found and is set as an amount of positional deviation of a channel with low reliability. Alternatively, plural pieces of positional deviation information with high reliability are subjected to processing, such as compilation and statistical processing, to uniquely calculate an amount of positional deviation common to all the channels to set the amount of positional deviation as an amount of positional deviation for each channel. Alternatively, it is also possible to interpolate an amount of positional deviation of a channel with low reliability from amounts of positional deviation of plural channels with high reliability by interpolation or extrapolation. The interpolation or extrapolation may be performed by linear interpolation, or distortion due to vibration or the like may be represented by curve approximation, such as spline approximation. Consequently, it becomes possible to perform alignment following distortion of an image due to vibration of the stage or the like, even in the case in which there is only a little pattern information for calculation of an amount of positional deviation of a channel.

Next, the amount of signal correction for adjusting for a deviation in brightness in the brightness correction section <b>109</b> is calculated for the aligned two images. As a cause of the deviation of brightness, there are factors such as (1) slight difference in the thickness among chips of a semiconductor wafer serving as an object to be inspected, (2) a difference in the sensitivity of each pixel of an image sensor, (3) a difference in the amount of accumulated light due to unevenness of the speed of movement of a stage, and (4) fluctuation in an amount of illumination light. Among the factors, whereas factor (1) occurs at random depending upon a pattern of the semiconductor wafer, factors (2), (3), and (4) occur linearly or in a belt shape on a detected image as characteristics inherent in the inspection apparatus.

Thus, in accordance with the present invention, first, a difference in the brightness, which occurs in a belt shape as in factors (2), (3), and (4), is adjusted at a specific high frequency, and a difference in the brightness, which occurs at random as in factor (1), is adjusted at a period different from the period (i.e., in a different processing area).

<figref idref="DRAWINGS">FIG. 8</figref> shows a flow of processing for the adjusting a difference in the brightness which occurs linearly or in a belt shape. First, with respect to a detected image or a reference image for which alignment has been performed, the amount of correction required for adjusting for an unevenness in the brightness (unevenness of brightness by a unit of row), which occurs in a direction perpendicular to the traveling direction of the stage, is calculated (step <b>81</b> in <figref idref="DRAWINGS">FIG. 8</figref>).

An example of a method of calculating the amount of correction of brightness at a highest frequency, that is, for each row, will be described below. First, as shown in <figref idref="DRAWINGS">FIG. 10A</figref>, with respect to an object row (M pixels x one row), an amount of correction for adjusting the brightness is calculated using a value of N rows (N pixels×N rows) in the vicinity including the row. As an example of a method of calculating the amount of correction, a statistical amount is calculated as indicated by expression (1) below.

<maths id="MATH-US-00001" num="00001"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><msub><mi>E</mi><mi>f</mi></msub><mo>=</mo><mrow><mfrac><mn>1</mn><mrow><mo>(</mo><mi>NxM</mi><mo>)</mo></mrow></mfrac><mo>⁢</mo><mrow><munder><mo>∑</mo><mi>N</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><munder><mo>∑</mo><mi>M</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mi>F</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>i</mi><mo>,</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mi>J</mi></mrow><mo>)</mo></mrow></mrow></mrow></mrow></mrow></mrow><mo>⁢</mo><mstyle><mtext></mtext></mstyle><mo>⁢</mo><mrow><msub><mi>E</mi><mi>G</mi></msub><mo>=</mo><mrow><mfrac><mn>1</mn><mrow><mo>(</mo><mi>NxM</mi><mo>)</mo></mrow></mfrac><mo>⁢</mo><mrow><munder><mo>∑</mo><mi>N</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><munder><mo>∑</mo><mi>M</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mi>G</mi><mo>⁢</mo><mrow><mo>(</mo><mrow><mi>i</mi><mo>,</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mi>j</mi></mrow><mo>)</mo></mrow></mrow></mrow></mrow></mrow></mrow><mo>⁢</mo><mstyle><mtext></mtext></mstyle><mo>⁢</mo><mrow><msub><mi>σ</mi><mi>f</mi></msub><mo>⁢</mo><msqrt><mrow><mfrac><mn>1</mn><mrow><mo>(</mo><mi>NxM</mi><mo>)</mo></mrow></mfrac><mo>⁢</mo><mrow><mo>{</mo><mrow><munder><mo>∑</mo><mi>N</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><munder><mo>∑</mo><mi>M</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><msup><mrow><mo>(</mo><mrow><mrow><mi>F</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>i</mi><mo>,</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mi>j</mi></mrow><mo>)</mo></mrow></mrow><mo>-</mo><msub><mi>E</mi><mi>f</mi></msub></mrow><mo>)</mo></mrow><mn>2</mn></msup></mrow></mrow><mo>}</mo></mrow></mrow></msqrt></mrow><mo>⁢</mo><mstyle><mtext></mtext></mstyle><mo>⁢</mo><mrow><msub><mi>σ</mi><mi>G</mi></msub><mo>⁢</mo><msqrt><mrow><mfrac><mn>1</mn><mrow><mo>(</mo><mi>NxM</mi><mo>)</mo></mrow></mfrac><mo>⁢</mo><mrow><mo>{</mo><mrow><munder><mo>∑</mo><mi>N</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><munder><mo>∑</mo><mi>M</mi></munder><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><msup><mrow><mo>(</mo><mrow><mrow><mi>G</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><mi>i</mi><mo>,</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mi>j</mi></mrow><mo>)</mo></mrow></mrow><mo>-</mo><msub><mi>E</mi><mi>G</mi></msub></mrow><mo>)</mo></mrow><mn>2</mn></msup></mrow></mrow><mo>}</mo></mrow></mrow></msqrt></mrow></mrow></mtd><mtd><mrow><mi>expression</mi><mo>⁢</mo><mstyle><mspace width="0.6em" height="0.6ex" /></mstyle><mo>⁢</mo><mrow><mo>(</mo><mn>1</mn><mo>)</mo></mrow></mrow></mtd></mtr></mtable></math></maths>

Here, F(i, j) and G(i,j) indicate the brightness of a detected image and a reference image after alignment in a position (i, j). Then, an amount of correction is calculated as indicated by expression (2) below. <br />gain_x=σ_F/σ_G<br />offset_x<i>=E</i>_f−gain·<i>E</i>_G  expression (2)

Correction of the object row is performed with respect to the reference image as indicated by expression (3) below. <br /><i>G</i>_x(<i>i,j</i>)=gain_x<i>·G</i>(<i>i,j</i>)+offset_x  expression (3)

Similarly, an amount of correction for adjusting unevenness in the brightness (unevenness of brightness by a unit of column), which occurs in parallel with the traveling direction of the stage, is calculated (step <b>82</b> in <figref idref="DRAWINGS">FIG. 8</figref>). First, as shown in <figref idref="DRAWINGS">FIG. 10B</figref>, an amount of correction for adjusting the brightness is calculated with respect to an object column (one column×L pixels) using a value of N columns (N columns×L pixels) in the vicinity including the column, and correction of the object column is performed with respect to the reference image as indicated by expression (4) below. <br /><i>Gy</i>(<i>i,j</i>)=gain_y<i>·G</i>(<i>i,j</i>)+offset_x  expression (4)

Then, as shown in <figref idref="DRAWINGS">FIG. 10C</figref> a logical product is found to calculate an amount of correction of a brightness of each pixel which occurs linearly or in a belt shape (step <b>83</b> in FIG. <b>8</b>).

If(F(i,j)−Gx(i,j))&lt;(F(i,j)−Gy(i,j))—Amount of correction=gainx, offsetx

Otherwise—Amount of correction=gainy, offsety

Next, an amount of correction for adjusting for a difference in the brightness, which occurs at random depending upon the pattern of a semiconductor wafer, is calculated at a frequency different from the amount of correction for the linear difference of brightness, that is, for each area of a different size.

As shown in the final difference image shown in <figref idref="DRAWINGS">FIG. 10C</figref>, a belt-like difference of brightness (noise), which occurs vertically or horizontally in a traveling direction of the stage, is reduced by finding a logical product of the respective pixels. However, the difference in the brightness which occurs at random remains without being corrected. Consequently, in accordance with the present invention, adjustment of the brightness targeting the random noise, which remains in this way, is performed next. An example of a procedure therefor is shown in <figref idref="DRAWINGS">FIGS. 17 and 18A</figref> to <b>18</b>C.

First, as shown in <figref idref="DRAWINGS">FIG. 18A to 18C</figref>, with respect to a detected image and a reference image for which a belt-like brightness is adjusted, a characteristic amount of each pixel in a specific area is calculated to form a characteristic space of two-dimensions or more, as shown in <figref idref="DRAWINGS">FIG. 18A</figref> (<b>17</b>-<b>1</b>). The characteristic amount maybe any amount as long as it indicates a characteristic of each pixel such as the contrast, the brightness, a second deviation value, a density difference between corresponding pixels, or a variance value using a near-field pixel. Next, the characteristic space is divided into plural segments (<b>17</b>-<b>2</b>), and an amount of correction is calculated for each segment using a statistical amount of pixels belonging to the segment (<b>17</b>-<b>3</b>). This is equivalent to, with respect to each pixel in areas of the detected image and the reference image, resolving a scatter diagram <b>18</b>B according to pixels in the areas, in which a brightness of the detected image is plotted on the X axis and a brightness of the reference image is plotted on the Y axis, as shown in <figref idref="DRAWINGS">FIGS. 18C and 18D</figref>, according to the characteristic amount, and an amount of correction is calculated in each scatter diagram.

<figref idref="DRAWINGS">FIGS. 19A and 19B</figref> show an example of a division method for dividing the characteristic space of <figref idref="DRAWINGS">FIG. 18A</figref> into segments. In accordance with the present invention, the segment division is automatically performed according to an object image. The upper graph in <figref idref="DRAWINGS">FIG. 19A</figref> shows an example of a characteristic space according to a brightness and a density difference, and the lower graph is a histogram showing the frequency of each brightness (luminance value), which is an example of determining a division threshold value in a brightness direction from a histogram of brightnesses of an object image.

<figref idref="DRAWINGS">FIG. 20</figref> is a flow diagram which shows an example of a processing procedure therefor. First, a histogram of luminance values in an object area is calculated (<b>20</b>-<i>b</i>). This may be calculated from the detected image or the reference image, or it may be calculated from an average value of two images. Next, the luminance histogram is smoothed to remove a small peak or the like (<b>20</b>-<b>2</b>), and a differential value of the smoothed histogram is calculated (<b>20</b>-<b>3</b>). Next, the differential value is checked from the side of lower brightness, and a luminance value at which the value changes to a positive value is set as a Start and a luminance value at which the value changes to negative value next is set as an End (<b>20</b>-<b>4</b>). Then, a luminance value at which the differential value is maximized in a range from Start to End is set as a threshold value for division (<b>20</b>-<b>5</b>).

Consequently, as shown in <figref idref="DRAWINGS">FIG. 19A</figref>, the division is performed in the part of the trough of the histogram. This indicates that the segment division is performed according to a pattern in the object area. It is possible to divide a luminance value according to the pattern in the image. It is also possible to divide a luminance value according to a fixed value set by a user as shown in <figref idref="DRAWINGS">FIG. 19B</figref>.

For the amount of correction of each scatter diagram resolved by the segment division, a linear equation is found by a least square approximation within the scatter diagram, as shown in <figref idref="DRAWINGS">FIGS. 18C and 18D</figref>, and an inclination and a y intercept are set as an amount of correction. In addition, an amount of correction may be found from pixels belonging to each segment, as shown in the above-mentioned expression <b>1</b> and expression <b>2</b>. Further, an area for forming a characteristic space can be set as an arbitrary area of pixels of 1×1 or more. However, since a defect is also adjusted if correction is performed with a 1×1 pixel of a highest frequency, the area is set so as to be slightly larger.

A result of the correction is shown in <figref idref="DRAWINGS">FIGS. 11A to 11E</figref>. <figref idref="DRAWINGS">FIG. 11A</figref> shows a difference image after alignment. A defect is in the encircled part. <figref idref="DRAWINGS">FIG. 11D</figref> shows a luminance waveform of two images after alignment of <b>1</b>D-<b>1</b>D′ including the defect part and a differential value at that point. The detected image has a defect, and the defect part is brighter than the part around it. However, the reference image is generally bright, and a differential value of the defect part is smaller than the part around it. On the other hand, <figref idref="DRAWINGS">FIG. 11B</figref> shows a difference image after correcting a difference of linear brightness at a high frequency. Belt-like unevenness of brightness is corrected. As a result, the luminance value is adjusted, as shown in <figref idref="DRAWINGS">FIG. 11E</figref>, and the defect is made manifest. However, this depends upon a repetitive pattern, and a difference in the brightness which occurs at random is not corrected.

<figref idref="DRAWINGS">FIG. 11C</figref> shows a difference image after calculating an amount of correction from a statistic amount of an area (in this example, the entire area of the image of <figref idref="DRAWINGS">FIG. 11B</figref>), which is different from the adjustment of a linear noise, and correcting a difference in the brightness. In this way, the amounts of correction are calculated in two or more different processing areas, whereby it becomes possible to correct the differences in brightness which occur in different circumstances, and to make manifest a defect of a feeble signal, which is embedded and invisible in a strong unevenness of the brightness; and to detect the defect.

In accordance with the present invention, with respect to the remaining random noise, adjustment of the brightness can be further performed for each area with similar characteristics. For example, when a characteristic space is formed in both a bright area, such as the peripheral circuit section <b>72</b> of <figref idref="DRAWINGS">FIG. 7</figref>, where a false defect easily occurs, and the memory mat section <b>71</b>, and an amount of correction is calculated, it is likely that a defect in the memory mat section <b>71</b> will be corrected. Thus, characteristic spaces are formed separately in the memory mat section <b>71</b> and the peripheral circuit section <b>72</b>. In addition, some memory mat sections or some peripheral circuit sections are in an area where noise easily occurs, and other memory mat sections or other peripheral circuit sections are in an area where noise does not easily occur. In such a case, characteristic spaces are also formed separately. The areas can be divided on the basis of design information of chips, such as CAD data, chip layout data, or chip images, or they can be divided on the basis of a result of a test inspection. In addition, an area forming one characteristic space may be discrete (discontinuous). A method of dividing an area may be described in detail later.

An effect of performing brightness correction of multiple stages with respect to an inspection image, as described above, will be described for the case of inspecting a pattern, which is formed on a semiconductor wafer covered by an insulating film with an optical transparent and flat surface, through, for example, a CMP process. Images which are obtained by sensing images of the wafer after the CMP processing in the detection unit <b>13</b> are affected by distribution of an amount of reflected light or the like, which is caused by fluctuation in thickness of the insulating film in the wafer surface and the irregularity of a pattern in a chip, so as to be changed to images with a brightness that fluctuation depending upon the location on the wafer. By performing brightness correction in multiple stages at different frequencies with respect to an image that is subject to fluctuation in brightness according to the method described above, the influence of the fluctuation in brightness among the images can be reduced to make the defect manifest. Thus, it becomes possible to improve the rate of detection of a defect.

In this way, after calculating the amount of correction for adjusting for a difference in the brightness using plural processing units, a comparison of the brightness in corresponding positions of two images is performed in the image comparison section <b>110</b> using an amount of positional deviation and amounts of signal correction which are calculated at two different periods. Then, a part where a differential value is larger than the threshold value TH is detected as a defect.

To summarize the flow of this processing, as shown in <figref idref="DRAWINGS">FIG. 9</figref>, in the image comparison processing unit <b>15</b>, first, an amount of positional deviation between a detected image and a reference image is calculated from an image signal, which is received from the image editing unit <b>14</b>, in the positional deviation correction section <b>108</b> (<b>1081</b>), and positional deviation of the detected image and the reference image is corrected on the basis of this calculated amount of deviation (<b>1082</b>). Next, in the brightness correction section <b>109</b>, an amount of correction of brightness unevenness between corresponding linear and belt-like areas of the detected image and the reference image is calculated (<b>1091</b>), and then an amount of correction of random brightness unevenness between corresponding areas of the detected image and the reference image is calculated for each area that is larger than the linear and belt-like areas (<b>1092</b>).

Next, in the image comparison section <b>110</b>, a differential image of the detected image and the reference image, which is corrected using information on the amount of positional deviation between the detected image and the reference image calculated in the positional deviation correction section <b>108</b>, information on the brightness unevenness between the linear and belt-like areas calculated in the brightness correction section <b>109</b>, and information on the amount of correction of random brightness unevenness calculated for each area larger than the linear and belt-like areas, is compared with the threshold value signal from the threshold value setting section <b>111</b> to detect a defect.

Here, in a semiconductor wafer which serves as an object of inspection, depending upon the position on a chip, there is an area which is not intentionally detected, even if the noise is large, a false defect easily occurs, or there is a defect in the area, because damage due to the noise, the false defect, or the defect is small. For example, in <figref idref="DRAWINGS">FIG. 7</figref>, the memory mat section <b>71</b> consists of a set of small repetitive patterns (cells), and the peripheral circuit section <b>72</b> is basically a set of random patterns. In general, in the memory mat section <b>71</b>, the pattern density is high and an image obtained by a bright-field illumination optical system tends to be dark. On the other hand, in the peripheral circuit section <b>72</b>, the pattern density is low and an image to be obtained tends to be bright. Further, in an area in which an image signal to be inputted is large (bright area), such as the peripheral circuit section <b>72</b>, since the level of noise is large and a false defect easily occurs, it is necessary to set the threshold value TH slightly larger. On the other hand, in the memory mat section <b>71</b>, since an image signal to be obtained is small (the section is dark), it is necessary to set the threshold value TH smaller and perform inspection with high sensitivity.

In addition, in the case in which a difference in the brightness occurs locally due to a difference in the thickness in a specific pattern area between chips as shown in <figref idref="DRAWINGS">FIGS. 4A and 4B</figref>, a false defect easily occurs in this part. Thus, it is necessary to set the threshold value TH slightly larger. On the other hand, in the case in which a comparison is performed using only the threshold value TH for the entire area, it is inevitable to set a slightly larger threshold value which is suited to an area with large noise or an area where a false defect easily occurs. Thus, the sensitivity in the other areas falls. Therefore, as a method of performing inspection with a high sensitivity over the entire area, the present invention provides means for setting the threshold value according to an area.

The characteristics <b>12</b>-<b>1</b> in <figref idref="DRAWINGS">FIG. 12</figref> indicates a threshold value for preventing a false defect from occurring in a certain position in a chip, indicating that it is necessary to set different threshold values according to an area. Moreover, in a semiconductor wafer, a difference in the brightness between chips tends to be larger in a part closer to the periphery of a wafer. The characteristic <b>12</b>-<b>2</b> in <figref idref="DRAWINGS">FIG. 12</figref> indicates a threshold value for preventing a false defect from! occurring in a certain position in the wafer, indicating that it is necessary to set different threshold values according to the position of a chip. In order to realize a highly sensitive inspection over the entire area of the inspection area in this way, in the threshold value setting section <b>111</b> of <figref idref="DRAWINGS">FIGS. 1 and 9</figref>, a threshold value is set for each position of the chip in the wafer and each position in the chip. An example of a procedure for effecting such setting is shown in <figref idref="DRAWINGS">FIG. 13</figref>.

First, a user sets general conditions (<b>13</b>-<b>1</b>) and performs a trial inspection (<b>13</b>-<b>2</b>). In this case, if only a threshold value according to a pattern in a chip is set, an area for one chip is set as an inspection area (<b>14</b>-<b>1</b> in <figref idref="DRAWINGS">FIG. 14</figref>). If a threshold value according to a position of a chip in a wafer is further set, chips on the entire surface of the wafer, or at least from an end to an end of the wafer, are set as an inspection area (<b>14</b>-<b>2</b> in <figref idref="DRAWINGS">FIG. 14</figref>). Then, the user visually confirms the presence of detected defect candidates and classifies them into real defects and false defects (<b>13</b>-<b>3</b>). For such confirmation, an image, which is sensed by image sensing means different from that used at the time of actual inspection, such as a high magnification camera, or an image used for the actual inspection, may be used. In this case, it is also possible to perform auto defect classification (ADC) or the like in parallel with the inspection and to automatically classify defect candidates into defects and false defects using a result of the auto defect classification or the like without intervention of the user. Then, a result of classification is displayed together with the chip, such that it can be seen at a glance where in the chip the occurrence of false defects are concentrated (<b>13</b>-<b>4</b>).

In the case in which plural chips are set as an inspection area, the results of inspection in the respective chips are superimposed. Looking at the superimposed results, the user sets an area where the user wishes to decrease the sensitivity, that is, an area where the threshold value is to be increased, and an area where the user wishes to increase sensitivity, that is, an area where the threshold value is to be decreased. In this case, as shown in <figref idref="DRAWINGS">FIG. 15</figref>, it is also possible to display the areas while being superimposed on design information of the chip, such as CAD data, chip layout data, and a chip image. By superimposition of display areas oh the design information, it becomes easier to confirm whether the set areas are appropriate and to change the set areas. In addition, an area can be set automatically from the design information of the chip.

Next, a characteristic amount of an area for setting a threshold value for each set area is calculated (<b>13</b>-<b>6</b>). For example, the characteristic amount is a characteristic amount F(i,j)−G(i,j), an area, or the like, of an area where a false defect occurs, such as a density difference value (a difference in brightness between the detected image and the reference image) of a false defect in an area where the occurrence of a false defect is concentrated, the pattern density, brightness, or the like of an area where a false defect occurs, and the like. In addition, in the case in which a trial inspection is performed in plural chips, the coordinates in a wafer of a chip where the occurrence of a false defect is concentrated, the shape of the occurrence of a false defect in the entire wafer, and the like are also calculated. Then, a threshold value, which is not a false defect, in each area is set from the characteristic amount in each area (<b>13</b>-<b>7</b>). For example, in the area where the occurrence of a false defect is concentrated, the detection of a false defect can be avoided if the threshold value is set to be larger than the density difference value of the false defect.

In addition, as described above, in an area where the pattern density is high or an area where the background is dark, the threshold value is set slightly lower because a false defect does not easily occur. On the other hand, in an area where the pattern density is low or an area where the background is bright, the threshold value is set to be slightly higher because a false defect easily occurs. Further, the threshold value is set slightly higher in a chip which is spaced from the center of a chip or a wafer where a false defect is concentrated. Since information such as the pattern density and the brightness of the background can be obtained from an image, division of an area is performed automatically.

As shown in <figref idref="DRAWINGS">FIG. 16A</figref>, the threshold value is basically set automatically from the density difference value of a false defect, but it may be set with reference to a threshold value set by the user in advance. In addition, as shown in <figref idref="DRAWINGS">FIG. 16B</figref>, a threshold value for judging detection or non-detection can also be calculated from the density difference value and the area. Naturally, it is also possible for the user to set the area and the threshold value for each area manually. Such automatic setting of the area and automatic threshold setting for each area are performed in section <b>111</b> in <figref idref="DRAWINGS">FIG. 1</figref>, and only a defect is detected in the image comparison section <b>110</b> using the threshold value.

Moreover, it is also possible to perform inspection with the threshold value determined in <b>13</b>-<b>7</b> in step <figref idref="DRAWINGS">FIG. 13</figref> and repeat steps <b>13</b>-<b>3</b> to <b>13</b>-<b>7</b> to perform tuning of the parameters. In addition, in this case, steps <b>13</b>-<b>3</b> to <b>13</b>-<b>7</b> can also be repeated while changing the optical conditions (focal point, amount of light, deflection conditions, etc.) (<b>13</b>-<b>8</b>) to perform tuning of the optical conditions.

As described above, in accordance with the present invention, image comparison processing is performed in parallel in the inspection for comparing two images and detecting a defect from a differential value thereof, whereby an inspection speed equivalent to or close to an image capturing speed of an image sensor can be realized.

For example, if the image capturing speed of the image sensor is 1.6 Gpps (pps: pixel per second) at the fastest, even if the processing capability of the image comparison processing unit <b>15</b> is only 0.8 Gpps, which is half the image capturing speed, it becomes possible to realize an inspection processing speed of 1.6 Gpps by adopting a constitution in which two image comparison processing units are provided in parallel, as in the present invention. Moreover, even in the case in which the speed of the image sensor is higher, such as the case in which the image capturing speed of the image sensor becomes equal to or higher than 1.6 Gpps, the case in which an image accumulation time of the image sensor is reduced according to optical conditions, or the case in which the scan width to be scanned by the image sensor at one time is increased, it becomes possible to cope with the case without speeding up the image comparison processing itself by increasing the number of image comparison processing units to N. For example, even in the case in which the image capturing speed of the image sensor is further increased to 6.4 Gpps at the fastest, it is possible to cope with a case having an image processing speed of 6.4 Gpps by arranging eight image comparison processing units <b>15</b> with a processing capability of 0.8 Gpps in parallel.

In addition, by adjusting the difference in the thickness among chips, a difference in the sensitivity of each pixel of the image sensor, a difference in the an amount of accumulated light due to unevenness of the stage speed, and a difference in the brightness among the chips which occurs due to various factors, such as fluctuation in illumination (unevenness of colors) at plural different frequencies (i.e., an amount of correction is calculated for each of plural different areas), it becomes possible to manifest a defect of a feeble signal, which is embedded in strong brightness unevenness, and detect the defect.

Further, by dividing an inspection area into plural areas according to coordinates in a wafer and coordinates in a chip to automatically set a threshold value for each divided area, the inspection sensitivity in each area can be optimized, so as to realize a highly sensitive inspection as a whole. In this case, by displaying design information of the chip, such as CAD data and a chip image of the chip so that it is superimposed on a state of occurrence of a false defect or a result of automatic area setting, confirmation and correction of a set area are facilitated. It is also possible for the user to set the threshold value manually.

The processing of the image comparison processing units <b>15</b>-<b>1</b> and <b>15</b>-<b>2</b> according to the present invention, as described above, is realized by software processing by the CPU. However, it is also possible to change a core arithmetic operation part, such as normalized cross correlation operation and the formation of a characteristic space, to hardware processing by an LSI. Consequently, further speed-up can be realized. In addition, even if there is a large difference in the brightness among compared dies due to a slight difference in the thickness of a pattern after a smoothing process, such as a CMP, or a reduction in the wavelength of the illumination light, it becomes possible to detect a defect of 20 nm to 90 nm according to the present invention.

Moreover, in inspection of a low k film, for example, an inorganic insulating film, such as SiO₂, SiOF, BSG, SiOB, or a porous silica film, or an organic insulating film, such as methyl containing SiO₂, MSQ, a polyimide film, a parelin film, a Teflon (registered trademark) film, or an amorphous carbon film, even if there is a local difference in the brightness due to in-film fluctuation of a refractive index distribution, it becomes possible to detect a defect of 20 nm to 90 nm according to the present invention.

An embodiment of the present invention has been described with reference to a comparative inspection image in an optical visual inspection apparatus targeting a semiconductor wafer as an example. However, the present invention is also applicable to a comparative image in an electron beam type pattern inspection. In addition, the object of inspection is not limited to a semiconductor wafer, and the present invention is applicable to a TFT substrate, a photomask, a printed board, or the like as long as a defect is detected by comparison of images.

As described above, according to the present invention, by performing image comparison processing in parallel, an inspection speed corresponding to a processing speed depending upon an image capturing speed of an image sensor, an image accumulation time, a scanning width, and the like can be realized.

In addition, by adjusting, at plural different frequencies, a difference of brightness among images to be compared caused by a difference in thickness among chips, a difference in the sensitivity of each pixel of the image sensor, a difference in the amount of accumulated light due to unevenness of the stage speed, a difference in the brightness among the chips which occurs due to various factors, such as fluctuation in illumination (unevenness of colors), or the like, it becomes possible to manifest a defect of a feeble signal, which is embedded in strong brightness unevenness, and detect the defect.

Further, since a threshold value can be set according to coordinates in a wafer and coordinates in a chip, it becomes possible to automatically optimize the inspection sensitivity in various places, so that highly sensitive inspection can be realized. In this case, by displaying design information of the chip and a threshold setting area so as to superimpose one on top of another, adjustment of the sensitivity, such as confirmation and correction of the threshold value setting area, is facilitated.

The invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiment is therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.